[rfc] Let the classloader search for system libraries, don't hard-code DEFAULT_LIBPATH

[Matthias Klose]

On 24.03.2011 15:43, Andrew Haley wrote:
> On 03/24/2011 02:18 PM, Matthias Klose wrote:
>> On 24.03.2011 15:09, Alan Bateman wrote:
>>> Matthias Klose wrote:
>>>> Currently os_linux.cpp defines DEFAULT_LIBPATH hard-coded to "/lib:/usr/lib";
>>>> with the upcoming multiarch changes in Debian/Ubuntu, libraries will move to
>>>> multiarch aware locations [1], where these libraries are not found anymore by
>>>> the class loaders.
>>>>   
>>> So is the issue that java.library.path isn't going to be set correctly when
>>> libraries are migrated to this new layout?
>>
>> Yes.
>>
>>> Just wondering if the setup of the
>>> search path (in os::init_system_properties_values would be better place to do
>>> consider this).
>>
>> DEFAULT_LIBPATH is only used in os::init_system_properties_values.
> 
> I'm trying to understand the problem.  Is it that the path to libraries is
> not fixed for a particular libjvm.so ?  I can't quite understand how this
> would be: surely if libjvm.so is built for ARM hard-FP then the libraries to
> search will be "/lib/arm-hardfp:/usr/lib/arm-hardfp" .  Or does the library
> path need to be dynamic?

No, afaics it doesn't have to be dynamic.